BAYAREALIKE or BAYAREALIKE + J

[Hannah H]

Hello all!

I have run BIOGEOBEARS Model test on my dataset with 105 tips and geographic areas assigned as A) Western Med B) Eastern Med C) Northern Europe D) Irano-Turania E) Far East Asia. Maximum areas assigned to one tip is 4. 

According to the out put, BAYAREALIKE is the best model with an AICc score of 426.1 and BAYAREALIKE+J is second best with a very similar AICc score of 428.3. LNL, d, e, and j parameters are similar if not the same. However, BAYAREALIKE does not seem to pick up on the same vicariance events that all other models do, including BAYAREALIKE + J. Why would this be? And given that the parameter scores between BAYAREALIKE and BAYAREALIKE + J are so similar, could I use BAYAREALIKE + J instead? I think the vicairance events that BAYAREALIKE do not pick up on are actually very plausible and explain the patterns I see and  provides a more biologically realistic interpretation.

Likelihood ratio test results: 

Log-Likelihoods = Both models have the same log-likelihood (-211), indicating equal fit.

Degrees of Freedom = BAYAREALIKE+J has one more degree of freedom than BAYAREALIKE (3 vs. 2).

Dstatistic = The likelihood ratio test statistic is -0.0084.

p-value = The p-value is 1, indicating no significant difference between the models.

Thank you for your help.

[Nick Matzke]

Hi,

Really, neither BAYAREALIKE nor BAYAREALIKE+J has any vicariance events.

The key is really understanding this chart:

http://phylo.wikidot.com/biogeobears#BioGeoBEARS_supermodel_graphic

The "looking at it by eye" interpretation of the output chart is not always a good indicator of what a particular model is assuming. The output chart is giving the best guess at each individual node/corner, but this is not always identical with a joint history.

BAYAREALIKE only ever copies the ranges to both daughters at speciation. All changes have to be range addition or range loss along branches.

BAYAREALIKE+J has that, plus jump dispersals at speciation.  These could be misinterpreted as vicariance to a new user, I suspect that is what's happening.

If you really think literal vicariance is happening, you should see this more clearly in DEC or DIVALIKE or the +J versions, particularly with stochastic mapping of the same.

But, which model fits best overall will depend on all the data in your tree, not just a few of the places with seemingly interesting events.  E.g., if your tree has lots of sister species with tip ranges like (AB, AB) or (ABC, ABCD), this will strongly support BAYAREALIKE as other models will require a number of extra events to explain those kinds of patterns.

The below post is similar and may help...

Cheers!

Nick

====================

Hi,

Assuming no errors in terms of naming models or whatever, a DIVALIKE model without (or even with) the +J model will not have a range-switching event (which is e.g. A->B event on a branch).

This does not, however, mean that you won't observe a history estimate like this:
 
       |-------B
----A|
       |-------A

This could be produced by, for example A->AB->B along a branch.  The first plot returned is a plot of "single most-probable range at each node or corner".  That plot represents the average of all possible histories probabilistically averaged according to the inferred parameters/rates.  It is just plotting the best-guess range at each node; the probability of that guess can be see in the pie-charts plot.

The most-probable range plot is not identical to "a proposed exact history of events that can be read off directly as a possible history".  For the latter, you need to do biogeographic stochastic mapping.

The two kinds of histories are often similar, but not always, and the "single most-probable range" plots can sometimes have apparent inconsistencies with the events allowed under a model.

Cheers!
Nick

====================

--
You received this message because you are subscribed to the Google Groups "BioGeoBEARS" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeobears...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeobears/128e5fa6-fe54-48b0-9c68-5b0d107678cfn%40googlegroups.com.

[Hannah H]

Hello,

Thank you so much for your speedy reply! I am new to this field, so I appreciate the help.

I have been running BioGeoBears in RASP. When looking at the output in "tree view", and selecting "vicariance" in highlight event, in both BAYAREALIKE and BAYAREALIKE+J, certain nodes are highlighted as vicariant (see attached). So, if BAYAREALIKE doesn't account for vicariance, what could these highlighted nodes mean instead? 

Best,

Hannah

[Nick Matzke]

Hi!  From memory, RASP (or some versions/parts of it) essentially add an extra inference on top of the original one -- probably basically a DIVA inference?  RASP was originally written to make the old DIVA program (Ronquist 1997) easier to use. Later, they added Lagrange DEC, BioGeoBEARS basic models, etc.

I suspect that in your graphic, RASP basically does this:

* If the most-probable ranges at each node/tip are:

    |------------| A

---|AB

    |------------| B

...then RASP concludes a vicariance event (AB -> A, B at speciation) happened.  DIVA or the DIVALIKE model of BioGeoBEARS would typically say this as well.

(An alternative possibility is that RASP is just literally running DIVA, as it is very fast, and placing that next to your BAYAREALIKE results)

However, the BAYAREALIKE model itself would actually propose histories like this:

    |-----(lose B)-------| A

---|AB

    |-----(lose A)-------| B

I am not the author of RASP so I am not sure, you would have to consult the documentation. 

Running the R script to get the plain BioGeoBEARS results is not too hard, so I would recommend doing that to get a clearer picture:

http://phylo.wikidot.com/biogeobears#script

Cheers!

Nick

To view this discussion on the web visit https://groups.google.com/d/msgid/biogeobears/c311a036-de8f-44ec-b05f-eec8f5d193e5n%40googlegroups.com.

[Nick Matzke]

On Mon, Jul 15, 2024 at 10:44 PM Hannah H <hhal...@gmail.com> wrote:

Hello,

Interestingly, when I run biogeobears in R (following the script provided), I get DEC as the best model! Which having some knowledge on the dispersal mechanisms of the group and their distribution throughout the Mediterranean and beyond, seems more likely. My next question is, how do I get the tree in an editable format?

R

If you mean "how do I get a graphics file I can edit", basically you take the PDF that the script gives, and/or save the screen graphic to PDF.  This PDF can then be edited in e.g. Adobe Illustrator or the free Inkscape.

Personally I like doing custom graphics all in R, BioGeoBEARS is just running graphics functions from the ape package to draw trees, you can see all the code if you type plot_BioGeoBEARS_results and hit return.  So you could do a custom graphic this way also (but it takes some code learning).

Cheers,

Nick

 

LnL numparams d e j AICc AICc_wt
DEC -215.7 2 0.022 1.0e-12 0 435.5 0.74
DEC+J -215.7 3 0.022 1.0e-12 1.0e-05 437.6 0.26
DIVALIKE -242.6 2 0.025 1.0e-12 0 489.2 1.6e-12
DIVALIKE+J -242.5 3 0.024 1.0e-12 0.0010 491.2 6.1e-13
BAYAREALIKE -229.7 2 0.017 0.040 0 463.5 6.3e-07
BAYAREALIKE+J -222.9 3 0.017 0.013 0.0094 452.1 0.0002

vs RASP

LnL numparams d e j AICc AICc_wt

DEC -224 2 0.030 0.065 0 452.1 1.7e-06

DEC+J -224 3 0.030 0.065 1.0e-05 454.3 5.7e-07

DIVALIKE -252.8 2 0.037 0.12 0 509.8 5.2e-19

DIVALIKE+J -252.8 3 0.038 0.14 1.0e-05 511.8 1.9e-19

BAYAREALIKE -211 2 0.024 0.16 0 426.1 0.74

BAYAREALIKE+J -211 3 0.024 0.16 1.0e-05 428.3 0.26

Best,

Hannah 

. 

To view this discussion on the web visit https://groups.google.com/d/msgid/biogeobears/6f16578e-b560-493f-a895-f8da7819d565n%40googlegroups.com.

[Hannah H]

Thank you for all your help! Much appreciated.

Best,

Hannah